Performance analysis of Machine learning approaches for developed real-time speaker’s voice recognition system

Speaker identification plays a vital role in various applications, including security systems, voice-activated devices, and personalized services, where real-time processing is key to ensuring dynamic and seamless user interactions. In this research, three machine learning (ML) classifiers— Convolutional Neural Networks (CNN), K-Nearest Neighbors (KNN), and Support Vector Machines (SVM)—are applied to create a real-time speaker recognition system. The dataset includes 160 audio files, composed of voices from direct recordings, with durations ranging from 2 to 8 seconds. Features like Mel-Frequency Cepstral Coefficients (MFCC) and Linear Predictive Coding (LPC) were extracted to train the models. The system categorizes input voices into two outputs: "Ayush" if the voice matches the speaker’s profile, and "Unrecognized" if it does not. Real-time validation was performed using live microphone input, with results displayed on an Arduino IDE serial-monitor and using an LED providing visual output. The system achieved swift identification, enhancing practical usability in real-time scenarios. Testing with 48 real-time voice samples revealed model accuracies of 91.67% for KNN, 97.92% for CNN, and 95.83% for SVM. This system demonstrates the application of ML classifiers for efficient speaker recognition and real-time voice identification in real-time operation.